Businesses can now build AI agents with the help of OpenAI’s new tools

OpenAI has introduced new tools aimed at helping developers and enterprises create AI agents-automated systems capable of performing tasks independently-using its advanced AI models and frameworks. Central to this release is the Responses API, which enables businesses to develop custom AI agents that can conduct web searches, analyze company files, and navigate websites, similar to OpenAI’s Operator product. This new API is set to replace the existing Assistants API, which OpenAI plans to phase out by mid-2026.

The concept of AI agents has gained significant attention in recent years, though the technology industry still faces challenges in defining and demonstrating their true capabilities. The recent buzz surrounding the Chinese startup Butterfly Effect and its Manus platform, which fell short of its promises, highlights the high expectations and challenges in this space. OpenAI is aware of these stakes. According to Olivier Godement, OpenAI’s API product head, while creating an agent is relatively straightforward, ensuring it operates at scale and sees regular usage is far more complex.

Earlier this year, OpenAI introduced two AI agents within ChatGPT: Operator, which assists with website navigation, and Deep Research, which compiles research reports. Though these tools offered a glimpse of agentic potential, they lacked a high level of autonomy. With the Responses API, OpenAI aims to provide developers with access to the core components behind such agents, allowing for the creation of more advanced and autonomous applications.

Developers using the Responses API can leverage the same AI models that power OpenAI’s ChatGPT Search tool, specifically the GPT-4o search and GPT-4o mini search models. These models are designed to browse the web and generate answers with cited sources. OpenAI claims high factual accuracy for these models, with GPT-4o search achieving a 90% score and GPT-4o mini search scoring 88% on the SimpleQA benchmark. For comparison, the larger GPT-4.5 model scores 63%.

The API also includes a file search feature that allows quick scanning of company databases to retrieve information, with assurances that OpenAI will not use this data for model training. Additionally, the API grants access to the Computer-Using Agent (CUA) model, which can automate tasks involving mouse and keyboard actions. While enterprises can choose to run the CUA model locally, the consumer version of Operator is limited to web-based actions.

However, OpenAI acknowledges that these tools won’t resolve all current challenges with AI agents. Despite improved accuracy, AI-powered search tools are still prone to errors and hallucinations, with GPT-4o search getting about 10% of factual queries wrong. The CUA model also remains unreliable for automating operating system tasks and is susceptible to mistakes.

Alongside the Responses API, OpenAI is launching the open-source Agents SDK, offering developers tools to integrate models with internal systems, establish safeguards, and monitor agent activities for optimization. This follows the earlier release of OpenAI’s Swarm framework for multi-agent orchestration.

Godement expressed hope that these developments will bridge the gap between AI agent demos and practical products, emphasizing that agents could become the most impactful AI application. This aligns with OpenAI CEO Sam Altman’s vision of 2025 as the year AI agents become integral to the workforce. While it remains to be seen whether AI agents will fully realize this potential in 2025, OpenAI’s latest initiatives indicate a strong push towards making agent technology more practical and impactful.

Author
Recent Posts

Pratik Patil

Pages

Businesses can now build AI agents with the help of OpenAI’s new tools

Categories

Pages

Categories

Headline